1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226

use std::arch::x86_64::{
    __m256i, _mm256_add_epi64, _mm256_add_epi8, _mm256_and_si256, _mm256_extract_epi64,
    _mm256_or_si256, _mm256_sad_epu8, _mm256_set1_epi8, _mm256_setr_epi8, _mm256_setzero_si256,
    _mm256_shuffle_epi8, _mm256_slli_epi64, _mm256_srli_epi32, _mm256_xor_si256,
};

use crate::distance_faster;

#[target_feature(enable = "avx2")]
unsafe fn carry_save_adder(h: *mut __m256i, l: *mut __m256i, a: __m256i, b: __m256i, c: __m256i) {
    let u = _mm256_xor_si256(a, b);
    *h = _mm256_or_si256(_mm256_and_si256(a, b), _mm256_and_si256(u, c));
    *l = _mm256_xor_si256(u, c);
}

#[target_feature(enable = "avx2")]
unsafe fn count(v: __m256i) -> __m256i {
    let lookup = _mm256_setr_epi8(
        0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4, 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3,
        3, 4,
    );
    let low_mask = _mm256_set1_epi8(0x0f);
    let lo = _mm256_and_si256(v, low_mask);
    let hi = _mm256_and_si256(_mm256_srli_epi32(v, 4), low_mask);
    let popcnt1 = _mm256_shuffle_epi8(lookup, lo);
    let popcnt2 = _mm256_shuffle_epi8(lookup, hi);
    let total = _mm256_add_epi8(popcnt1, popcnt2);
    _mm256_sad_epu8(total, _mm256_setzero_si256())
}

#[target_feature(enable = "avx2")]
unsafe fn avx2_harvey_seal_popcnt(x: *const __m256i, size: usize) -> u64 {
    let mut total = _mm256_setzero_si256();
    let mut ones = _mm256_setzero_si256();
    let mut twos = _mm256_setzero_si256();
    let mut fours = _mm256_setzero_si256();
    let mut eights = _mm256_setzero_si256();
    let mut sixteens = _mm256_setzero_si256();
    let mut twos_a = _mm256_setzero_si256();
    let mut fours_a = _mm256_setzero_si256();
    let mut eights_a = _mm256_setzero_si256();
    let mut twos_b = _mm256_setzero_si256();
    let mut fours_b = _mm256_setzero_si256();
    let mut eights_b = _mm256_setzero_si256();
    let mut i = 0;
    while i < size {
        carry_save_adder(&mut twos_a, &mut ones, ones, *x.add(i), *x.add(i + 1));
        carry_save_adder(&mut twos_b, &mut ones, ones, *x.add(i + 2), *x.add(i + 3));
        carry_save_adder(&mut fours_a, &mut twos, twos, twos_a, twos_b);
        carry_save_adder(&mut twos_a, &mut ones, ones, *x.add(i + 4), *x.add(i + 5));
        carry_save_adder(&mut twos_b, &mut ones, ones, *x.add(i + 6), *x.add(i + 7));
        carry_save_adder(&mut fours_b, &mut twos, twos, twos_a, twos_b);
        carry_save_adder(&mut eights_a, &mut fours, fours, fours_a, fours_b);
        carry_save_adder(&mut twos_a, &mut ones, ones, *x.add(i + 8), *x.add(i + 9));
        carry_save_adder(&mut twos_b, &mut ones, ones, *x.add(i + 10), *x.add(i + 11));
        carry_save_adder(&mut fours_a, &mut twos, twos, twos_a, twos_b);
        carry_save_adder(&mut twos_a, &mut ones, ones, *x.add(i + 12), *x.add(i + 13));
        carry_save_adder(&mut twos_b, &mut ones, ones, *x.add(i + 14), *x.add(i + 15));
        carry_save_adder(&mut fours_b, &mut twos, twos, twos_a, twos_b);
        carry_save_adder(&mut eights_b, &mut fours, fours, fours_a, fours_b);
        carry_save_adder(&mut sixteens, &mut eights, eights, eights_a, eights_b);
        total = _mm256_add_epi64(total, count(sixteens));
        i += 16;
    }
    // final reduce
    total = _mm256_slli_epi64(total, 4);
    total = _mm256_add_epi64(total, _mm256_slli_epi64(count(eights), 3));
    total = _mm256_add_epi64(total, _mm256_slli_epi64(count(fours), 2));
    total = _mm256_add_epi64(total, _mm256_slli_epi64(count(twos), 1));
    total = _mm256_add_epi64(total, count(ones));
    (_mm256_extract_epi64(total, 0)
        + _mm256_extract_epi64(total, 1)
        + _mm256_extract_epi64(total, 2)
        + _mm256_extract_epi64(total, 3)) as u64
}

#[target_feature(enable = "avx2")]
unsafe fn avx2_harvey_seal_dist(x: *const __m256i, y: *const __m256i, size: usize) -> u64 {
    let mut total = _mm256_setzero_si256();
    let mut ones = _mm256_setzero_si256();
    let mut twos = _mm256_setzero_si256();
    let mut fours = _mm256_setzero_si256();
    let mut eights = _mm256_setzero_si256();
    let mut sixteens = _mm256_setzero_si256();
    let mut twos_a = _mm256_setzero_si256();
    let mut fours_a = _mm256_setzero_si256();
    let mut eights_a = _mm256_setzero_si256();
    let mut twos_b = _mm256_setzero_si256();
    let mut fours_b = _mm256_setzero_si256();
    let mut eights_b = _mm256_setzero_si256();
    let mut i = 0;
    while i < size {
        carry_save_adder(
            &mut twos_a,
            &mut ones,
            ones,
            _mm256_xor_si256(*x.add(i), *y.add(i)),
            _mm256_xor_si256(*x.add(i + 1), *y.add(i + 1)),
        );
        carry_save_adder(
            &mut twos_b,
            &mut ones,
            ones,
            _mm256_xor_si256(*x.add(i + 2), *y.add(i + 2)),
            _mm256_xor_si256(*x.add(i + 3), *y.add(i + 3)),
        );
        carry_save_adder(&mut fours_a, &mut twos, twos, twos_a, twos_b);
        carry_save_adder(
            &mut twos_a,
            &mut ones,
            ones,
            _mm256_xor_si256(*x.add(i + 4), *y.add(i + 4)),
            _mm256_xor_si256(*x.add(i + 5), *y.add(i + 5)),
        );
        carry_save_adder(
            &mut twos_b,
            &mut ones,
            ones,
            _mm256_xor_si256(*x.add(i + 6), *y.add(i + 6)),
            _mm256_xor_si256(*x.add(i + 7), *y.add(i + 7)),
        );
        carry_save_adder(&mut fours_b, &mut twos, twos, twos_a, twos_b);
        carry_save_adder(&mut eights_a, &mut fours, fours, fours_a, fours_b);
        carry_save_adder(
            &mut twos_a,
            &mut ones,
            ones,
            _mm256_xor_si256(*x.add(i + 8), *y.add(i + 8)),
            _mm256_xor_si256(*x.add(i + 9), *y.add(i + 9)),
        );
        carry_save_adder(
            &mut twos_b,
            &mut ones,
            ones,
            _mm256_xor_si256(*x.add(i + 10), *y.add(i + 10)),
            _mm256_xor_si256(*x.add(i + 11), *y.add(i + 11)),
        );
        carry_save_adder(&mut fours_a, &mut twos, twos, twos_a, twos_b);
        carry_save_adder(
            &mut twos_a,
            &mut ones,
            ones,
            _mm256_xor_si256(*x.add(i + 12), *y.add(i + 12)),
            _mm256_xor_si256(*x.add(i + 13), *y.add(i + 13)),
        );
        carry_save_adder(
            &mut twos_b,
            &mut ones,
            ones,
            _mm256_xor_si256(*x.add(i + 14), *y.add(i + 14)),
            _mm256_xor_si256(*x.add(i + 15), *y.add(i + 15)),
        );
        carry_save_adder(&mut fours_b, &mut twos, twos, twos_a, twos_b);
        carry_save_adder(&mut eights_b, &mut fours, fours, fours_a, fours_b);
        carry_save_adder(&mut sixteens, &mut eights, eights, eights_a, eights_b);
        total = _mm256_add_epi64(total, count(sixteens));
        i += 16;
    }
    // final reduce
    total = _mm256_slli_epi64(total, 4);
    total = _mm256_add_epi64(total, _mm256_slli_epi64(count(eights), 3));
    total = _mm256_add_epi64(total, _mm256_slli_epi64(count(fours), 2));
    total = _mm256_add_epi64(total, _mm256_slli_epi64(count(twos), 1));
    total = _mm256_add_epi64(total, count(ones));
    (_mm256_extract_epi64(total, 0)
        + _mm256_extract_epi64(total, 1)
        + _mm256_extract_epi64(total, 2)
        + _mm256_extract_epi64(total, 3)) as u64
}

#[target_feature(enable = "avx2")]
pub unsafe fn distance_vect(x: &[u8], y: &[u8]) -> u64 {
    assert_eq!(x.len(), y.len());
    let mut accum = 0;
    let (x_head, x_mid, x_tail) = x.align_to::<__m256i>();
    let (y_head, y_mid, y_tail) = y.align_to::<__m256i>();
    if x_head.len() != y_head.len() {
        // TODO : warn
        return distance_faster(x, y);
    }
    accum += super::distance_faster(x_head, y_head);
    let main_block_length = 16 * (x_mid.len() / 16);
    let x_ptr_avx = x_mid.as_ptr() as *const __m256i;
    let y_ptr_avx = y_mid.as_ptr() as *const __m256i;

    accum += avx2_harvey_seal_dist(x_ptr_avx, y_ptr_avx, main_block_length);

    let x_avx_tail = x_ptr_avx.add(main_block_length) as *const u8;
    let y_avx_tail = y_ptr_avx.add(main_block_length) as *const u8;

    let x_final = std::slice::from_raw_parts(
        x_avx_tail,
        32 * (x_mid.len() - main_block_length) + x_tail.len(),
    );

    let y_final = std::slice::from_raw_parts(
        y_avx_tail,
        32 * (y_mid.len() - main_block_length) + y_tail.len(),
    );

    accum += super::distance_faster(x_final, y_final);

    accum
}

#[target_feature(enable = "avx2")]
pub unsafe fn weight_vect(x: &[u8]) -> u64 {
    let mut accum = 0;
    let (x_head, x_mid, x_tail) = x.align_to::<__m256i>();
    accum += super::weight_naive(x_head);
    let main_block_length = 16 * (x_mid.len() / 16);
    let x_ptr_avx = x_mid.as_ptr() as *const __m256i;

    accum += avx2_harvey_seal_popcnt(x_ptr_avx, main_block_length);

    let x_avx_tail = x_ptr_avx.add(main_block_length) as *const u8;

    let x_final = std::slice::from_raw_parts(
        x_avx_tail,
        32 * (x_mid.len() - main_block_length) + x_tail.len(),
    );

    accum += super::weight_naive(x_final);

    accum
}